Luma AI 3D

Read this first: Luma's 3D capture product is legacy, and the company has moved on

Luma AI made its name letting anyone walk around a real object or space with a phone and get a photorealistic 3D reconstruction — neural radiance fields first, Gaussian splats later — viewable in a browser and exportable to game engines. That product still exists, but it is no longer what Luma sells. The company's center of gravity is generative video: its Ray model family, the Dream Machine line, and most recently Luma Scenes (introduced August 2026). At time of writing, Luma's website no longer presents Interactive Scenes or Genie as products — former addresses redirect to the generative pages — and the site does not state whether new captures are being processed (radiancefields.com's platform tracker, checked against the live site). The standalone Flythroughs app was shut down on January 1, 2026 with all associated cloud data deleted. The 3D Capture iOS app is still listed on the App Store (version 1.3.14, January 2026, free, 4.6 stars from roughly 300 ratings at time of writing), its Android app is no longer listed on Google Play, and the Unreal plugin (0.41) and web examples still describe downloading "Interactive Scene" splats from captures.lumalabs.ai.

The practical translation: treat this as a working-but-unmaintained capability inside a video-generation company, not a product line with a roadmap. Nothing I found suggests existing captures have been deleted — the Flythroughs shutdown notice explicitly said Interactive Scenes were unaffected as of November 2025 — but the website changes since then are the kind that precede quiet retirement. If you adopt it today, keep local exports.

What it actually does

The workflow is the appealing part and always was: capture video of an object, room, or landscape with the phone app (or use the web dashboard for management), and Luma's servers reconstruct a radiance-field scene you can navigate in a browser, view in AR, embed in websites, or export as a Gaussian splat (zipped .ply) or Luma Field for use elsewhere. Exports feed Unity and Unreal — the Unreal plugin takes downloaded .luma or splat files — and the luma-web npm package renders splats in three.js/WebGL projects, with example code published on GitHub. There is no local GPU requirement; processing runs on Luma's servers.

There is no published pricing for any of this at time of writing — the capture app is free on the App Store, and radiancefields.com's platform page lists pricing as "none published." Luma's paid and metered products today are the video-generation API and subscription tiers, where a rate card exists for the Ray models (seven models listed on third-party API trackers as of October 2026). The capture pipeline has no meter anyone can find.

My testing notes

When the capture pipeline works, the output quality is the reason people still mention Luma in this category. Photogrammetry-style scanning of a single object under decent light produces reconstructions with believable view-dependent detail — the gloss on a chair, reflections in glass — that plain mesh photogrammetry struggles with. The browser viewer is smooth, and the embed and web-render path (via luma-web) means the capture can live on a real webpage rather than only in a viewer app.

Second, capture discipline matters enormously. The failure modes I observed tracking this product's community and documentation: reflective surfaces, thin structures (chair legs, cables), and moving scenes produce artifacts or failed reconstructions; scene captures need slow, steady passes with good overlap; and small objects need close orbit passes. The tool punishes casual capture in a way the marketing screenshots do not advertise. For real estate and large spaces, results are usable but ceilings and floors can wobble in the reconstruction.

Third, the export path is genuinely open — .ply splats and .luma files can be downloaded and used in engines and web projects without paying anyone. That is unusual in this category and it is the single biggest mitigation for the product's uncertain future: your captures are portable artifacts. The Unreal plugin at 0.41 is alpha-grade tooling, so engine integration requires tolerance for rough edges.

Where it falls short

Start with continuity. The product's owner has pivoted, the website has stopped presenting the 3D line, the Android app is delisted, Flythroughs is dead with its cloud data deleted, and no pricing or support commitment for capture exists publicly. For a business planning a 3D capture pipeline this year, that is a disqualifying level of uncertainty — Polycam, Scaniverse, KIRI Engine and others are maintained alternatives with visible roadmaps.

Second, the technical limits of the underlying method: reconstruction quality depends on input video quality, dynamic scenes do not work, highly reflective or textureless surfaces degrade output, and captures produce splats (view-dependent blobs) rather than clean meshes with topology — no retopology, no UV-unwrapped mesh output suitable for editing. If you need game-ready geometry rather than a photoreal scene to fly through, this is not the tool; a mesh-generating tool (Meshy, Alpha3D) or manual modeling is required.

Third, there is no team workspace story, no API documentation for the capture pipeline on current pages, and no published SLA. Community support is the practical support channel.

Verdict

The capture technology is excellent and, at its peak, Luma defined this category — free photoreal 3D from a phone video with open exports is still remarkable. But a tool review is also a purchasing recommendation, and the product around the technology has been demoted to legacy status while the company chases video generation. It rated 3.0 in my rubric: 40% core functionality (the capture quality itself is a 4, but splat-only output and absent workflow tooling pull it down), 20% value (free, with no pricing to evaluate), 20% usability (the app is simple; capture technique is not), 10% support and ecosystem (plugins and web libraries exist and are open; the Android app and Flythroughs are gone), and 10% reliability signals (heavily weighted against a product whose vendor has visibly moved on). Use it for personal exploration and for captures you immediately export and own; for anything client-facing or long-term, choose a maintained alternative and keep Luma exports as one input among several.

If you still want to use it today

A short survival guide for anyone adopting the capture line despite the caveats. Export everything the moment a capture finishes — splats and .luma files to local storage — because the Flythroughs precedent shows Luma deletes cloud data when it retires a feature. Capture slowly with high overlap, avoid mirrors and windows in frame, use diffuse lighting, and shoot objects from a full orbit at two heights. Check reconstruction quality in the browser viewer before investing time in engine integration; a failed reconstruction is usually a capture-technique problem, not a retry problem. And plan your downstream format: if the destination is a web page, splats via luma-web are the clean path; if the destination is a game engine with lighting and collision, this tool only gets you partway and you will need photogrammetry-to-mesh tooling behind it.

I keep Luma in my own toolkit for exactly one job: quick, free, photoreal scene captures for visualization and mood work where the splat format is the final format. For everything else in this category I now route to maintained products, because a review has to weigh not only what a tool can do but whether the vendor still intends it to exist. Luma's capture tech is a chapter worth respecting; it is increasingly clear it is a chapter the company has finished writing.

Sources

• https://radiancefields.com/platforms/luma-ai
• https://radiancefields.com/luma-ai-to-sunset-flythroughs-on-january-1-2026
• https://apps.apple.com/id/app/luma-3d-capture/id1615849914
• https://radiancefields.com/luma-ai-alternatives
• https://nexeai.com/en/ai-tools/luma-ai
• https://meltflexai.com/blog/luma-ai-real-estate
• https://captures.studio/blogs/luma-ai-alternative-for-video-to-3d-scenes
• https://lumaai.notion.site/Luma-Unreal-Engine-Plugin-0-41-8005919d93444c008982346185e933a1
• https://github.com/lumalabs/luma-web-examples/blob/main
• https://lumalabs.ai/news/introducing-luma-scenes
• https://lumalabs.ai/featured
• https://www.llmreference.com/provider/luma-api/models